During radiation therapy for prostate cancer it is common practice to localize the prostate on many, ideally all, days of treatment to achieve the goal of delivering a high dose to the prostate while greatly sparing nearby radiosensitive normal tissues. The prostate can be localized in CT images acquired immediately prior to treatment or by tracking markers, implanted in the prostate, during treatment delivery. The advantage of CT imaging is that the image data provides the basis for computing radiation doses actually delivered to the prostate and surrounding tissues as needed for Adaptive Radiation Therapy (ART), a procedure for periodically adjusting the treatment plan in order to deliver a final dose distribution as originally planned. Disadvantages of CT imaging include monetary, space and time expenses. Major advantages of marker tracking include ease of use and frequent sampling during each dose fraction, e.g., 10 Hz, potentially allowing dynamic adjustment of treatment parameters. Unfortunately the absence of image data from tracking systems prevents the practice of ART, and for some patients may preclude the use of smaller margins around the prostate that would allow better sparing of nearby normal tissues. The hypothesis of the proposed research is that the prostate markers localized during treatment delivery can be used as the basis for mapping reference CT image data into the treatment space to estimate pre-treatment CT image data acceptable for calculating delivered dose. In particular the markers will be used for computing the image-match term in a Bayesian-like framework to optimize the non-rigid registration of a statistically trainable deformable shape model of a prostate, including immediately surrounding tissues, with marker positions in the treatment space. A patient-specific model, called an m-rep, created from the reference planning image for the patient being treated will embed the underlying image data, including prostate-related marker coordinates, in the model-related coordinate system unique to m- reps. The deformed m-rep created by the registration process to match the markers located during treatment implies a transformation that maps the entire reference image data to the treatment space to estimate the pre-treatment CT image. The tissue region in and around the prostate is mapped diffeomorphically. The overall aim is to establish proof of concept for estimating pre-treatment CT images acceptable for calculating delivered dose as described above PUBLIC HEALTH RELEVANCE: The hypothesis of the proposed research is that prostate markers localized during radiation therapy treatment delivery can be used as the basis for mapping reference CT image data into the treatment space to estimate pre-treatment CT image data acceptable for calculating delivered dose. In particular the markers will be used for computing the image-match term in a Bayesian-like framework to optimize the non-rigid registration of a statistically trainable deformable shape model of a prostate, including immediately surrounding tissues, with marker positions in the treatment space. A patient- specific model, called an m-rep, created from the reference planning image for the patient being treated will embed the underlying image data, including prostate-related marker coordinates, in the model-related coordinate system unique to m-reps. The deformed m-rep created by the registration process to match the markers located during treatment implies a transformation that maps the entire reference image data to the treatment space to estimate the pre-treatment CT image. The tissue region in and around the prostate is mapped diffeomorphically. The overall aim is to establish proof of concept for estimating pre-treatment CT images acceptable for calculating delivered dose as described above.